Selected work across machine learning, causal inference, LLM evaluation, computer vision, and software engineering.
Causal Inference / Uplift Modeling
Apr 2026 — Aug 2026
An end-to-end uplift modeling system built on a 200K-customer randomized retail experiment. The system constructs 47 leakage-safe features from 45.8M transaction records, compares causal targeting against propensity modeling, and includes a budget-constrained targeting optimizer with a Streamlit simulator.
LLM Evaluation / Semantic Consistency
Jul 2026 — Present
A semantic self-consistency benchmark for evaluating whether large language models preserve meaning under controlled perturbations. The framework combines symbolic verification, exact matching, NLI, and LLM-as-judge evaluation with automated probe generation and scalable GPU inference.
Machine Learning / Explainable AI
Jan 2026 — May 2026
A multimodal disease-diagnosis system pairing a LoRA fine-tuned vision-language model with an EfficientNet-B0 + Grad-CAM explainability pipeline to produce human-verifiable visual and textual explanations. Submitted to the ICML 2026 AI4GOOD Workshop.
Full-Stack Online Judge
May 2026 — Jul 2026
A full-stack, LeetCode-style online coding judge that compiles, executes, and evaluates C++ submissions against hidden test cases through an automated verdict pipeline (Accepted, Wrong Answer, TLE, Runtime & Compilation errors), with an optional Docker sandbox for isolated execution. Deployed across Vercel, Render, and Neon.
Distributed Systems
Sep 2025 — Dec 2025
A fault-tolerant distributed booking system built on the Raft consensus algorithm for leader election and log replication, with six gRPC microservices communicating over Protocol Buffers and concurrent seat reservation that prevents double booking under simultaneous requests.